WEBVTT - Bloomberg Technology Special: Nvidia CEO Jensen Huang 0:00:01.480 --> 0:00:05.320 We're from Mahart. We're Innovation, Money and Power Collie in 0:00:05.440 --> 0:00:10.560 Silicon Vallet NBN. This is Bloomberg Technology with Caroline Hyde and. 0:00:10.720 --> 0:00:11.800 Ed Ludlow. 0:00:25.960 --> 0:00:29.320 Live from San Francisco to our TV and radio audiences 0:00:29.360 --> 0:00:33.320 around the world. Welcome to a special edition of Bloomberg Technology. 0:00:33.320 --> 0:00:35.920 I'm Ed Ludlow. In just a few moments in vidio, 0:00:36.000 --> 0:00:39.320 CEO Jensen Wang will join us for a live interview 0:00:39.400 --> 0:00:43.040 following their latest earnings report, the company posting a revenue 0:00:43.040 --> 0:00:46.280 forecast that beat consensus that fell short to some of 0:00:46.320 --> 0:00:51.360 the most optimistic estimates, stoking concern that the explosive growth 0:00:51.800 --> 0:00:55.200 is waning. Let's get right to Bloomberg Semiconductor correspondent Ian King, 0:00:55.240 --> 0:00:57.840 who joins me on set. Let's start with the basics, 0:00:57.880 --> 0:01:00.760 the fiscal third quarter forecast and we learned through it. 0:01:00.920 --> 0:01:04.680 Yeah, that forecast was fine. If you compare it to consensus, 0:01:04.680 --> 0:01:07.240 it was there or thereabouts. And for most of the 0:01:07.319 --> 0:01:09.720 companies in the world, that would be great and everybody 0:01:09.760 --> 0:01:11.920 would be happy. But this is in video. This is 0:01:11.959 --> 0:01:15.840 a company which you know, beats pretty much every quarter 0:01:16.160 --> 0:01:19.160 and by an order of magnitude, and it didn't do that, 0:01:19.240 --> 0:01:21.200 and it didn't indicate it would do that, and that 0:01:21.280 --> 0:01:23.119 raised a lot of questions. As we heard on the. 0:01:23.040 --> 0:01:25.680 Call, there are many storylines. I think the demand from 0:01:25.680 --> 0:01:29.039 the hyperscalers is clearly intact, but Blackwell was everything. I 0:01:29.080 --> 0:01:32.039 want to play a SoundBite of what Jensen Wong said 0:01:32.200 --> 0:01:33.480 about Blackwell. Listened to this. 0:01:34.520 --> 0:01:38.640 The change to the mask is complete. There were no 0:01:38.959 --> 0:01:46.880 functional changes necessary, and so we're sampling functional samples of 0:01:47.640 --> 0:01:51.880 Blackwell Grace Blackwell in a variety of system configurations as 0:01:51.920 --> 0:01:52.360 we speak. 0:01:53.960 --> 0:01:56.400 The main point here is that there was not a 0:01:56.480 --> 0:02:00.520 design issue with Blackwell itself as had been reported, but 0:02:00.680 --> 0:02:03.800 based on what Nvidia said, this was about the production mechanism. 0:02:04.000 --> 0:02:06.840 Blue Bazine King, you did a very good job in 0:02:06.880 --> 0:02:10.320 the top Life blog of explaining a GPU mask. Could 0:02:10.320 --> 0:02:12.720 you just try to give a short version of that 0:02:12.760 --> 0:02:13.400 to our audience. 0:02:13.440 --> 0:02:17.079 Yeah, the mask is basically the blueprint, which basically that 0:02:17.560 --> 0:02:20.280 is used to burn in the circuit onto the surface 0:02:20.320 --> 0:02:22.280 of the chip to give it its function, right, and 0:02:22.360 --> 0:02:25.280 it is a very important step. That's the fundamental blueprint. 0:02:25.639 --> 0:02:29.240 And they were saying we didn't get it wrong, but 0:02:29.440 --> 0:02:31.920 when it came to manufacturing, it didn't produce as many 0:02:31.960 --> 0:02:34.360 good chips as we wanted, so we made some tweaks 0:02:34.360 --> 0:02:37.000 to that. Didn't have to redesign it, but made some tweaks, 0:02:37.280 --> 0:02:39.120 and that is helping us to get a better yield. 0:02:39.400 --> 0:02:42.480 It's worth noting at this stage the stocks down almost 0:02:42.520 --> 0:02:44.640 seven percent in after hours, have been down more at 0:02:44.680 --> 0:02:47.360 the conclusion of the call. I think basically because we 0:02:47.360 --> 0:02:50.560 didn't learn enough about Blackwell. What they said was in 0:02:50.600 --> 0:02:54.040 the fiscal fourth quarter of fiscal year twenty five, there 0:02:54.040 --> 0:02:58.840 will be several billion dollars of sales through Blackwell. Why 0:02:58.880 --> 0:03:00.960 does the market want more and what does it want? 0:03:01.200 --> 0:03:04.800 They wanted a reassurance from in videos management, and they 0:03:04.840 --> 0:03:09.080 wanted reassurance in the form of details. They wanted gens 0:03:09.120 --> 0:03:11.520 to PuTTY's arm around everybody and say, don't worry, It'll 0:03:11.520 --> 0:03:13.480 be fine. This is how much I'm going to get. 0:03:13.720 --> 0:03:17.040 They asked consistently, We're asked questions of how many billions 0:03:17.160 --> 0:03:20.800 and when will those billions exactly come in and collect. Kretz, 0:03:21.080 --> 0:03:25.200 the CFO and Jensen Wang essentially avoided that question and 0:03:25.360 --> 0:03:27.480 refuse to give that precise reassurance. 0:03:28.600 --> 0:03:30.800 We're showing some of the after ours reaction, not just 0:03:30.840 --> 0:03:32.919 in in video itself, but some of its peers, both 0:03:32.960 --> 0:03:36.760 on the chip making side the server equipment providers, and 0:03:37.080 --> 0:03:39.960 I think AMD and ARM in particular are very noteworthy. 0:03:40.280 --> 0:03:41.960 Bloombogty and King stay with us. I want to go 0:03:42.000 --> 0:03:44.680 out to Chicago and Bloomberg's Ryan for Lostelka on I 0:03:44.760 --> 0:03:47.840 Equities team and Ryan. That's the broad summary from me 0:03:48.080 --> 0:03:50.680 about the names moving and after hours there's probably a 0:03:50.680 --> 0:03:53.280 bigger picture after ours movement in the markets as well, 0:03:53.480 --> 0:03:55.839 start with in video and work outward from there. 0:03:56.640 --> 0:03:58.960 Sure, well, one thing I would say about in videos 0:03:59.000 --> 0:04:01.440 after hours decline is that it does come after a 0:04:01.600 --> 0:04:04.000 very strong year to day performance. I think it closed 0:04:04.040 --> 0:04:05.960 up more than one hundred and fifty percent this year, 0:04:06.280 --> 0:04:09.120 So even though the re forecast was maybe a little 0:04:09.160 --> 0:04:11.800 bit shy of some of the most optimistic expectations, it 0:04:11.800 --> 0:04:14.600 did beat expectations. And it is, you know, coming off 0:04:14.640 --> 0:04:17.080 such a huge gain. So it's not necessarily surprising to 0:04:17.120 --> 0:04:19.720 see a little bit of a consolidation now. And I 0:04:19.760 --> 0:04:21.600 think even the declient is a little bit less and 0:04:21.640 --> 0:04:24.960 the options market was anticipating. So just some context there 0:04:24.960 --> 0:04:27.159 for looking at the declient. But you're right that pretty 0:04:27.200 --> 0:04:30.960 much we are seeing widespread weakness following this report, all 0:04:30.960 --> 0:04:34.160 the megacaps are, you know, modestly lower. We are seeing 0:04:34.240 --> 0:04:37.080 much more pronounced weakness in other chip makers and chip 0:04:37.120 --> 0:04:39.960 design companies and so forth. So yeah, certainly it does 0:04:40.000 --> 0:04:42.320 seem like the initial read through here is negative, but 0:04:42.360 --> 0:04:44.640 again it does come after a very strong start to 0:04:44.680 --> 0:04:45.000 the year. 0:04:46.080 --> 0:04:48.680 We made a big deal about this earnings print for 0:04:48.800 --> 0:04:51.640 quite a long time. I look at something like an 0:04:51.640 --> 0:04:55.800 investco QQQ, the ETF that tracks the nast that one hundred, 0:04:56.120 --> 0:04:58.479 and I think it's down around a percentage point, right, 0:04:58.560 --> 0:05:02.599 But in true in the market's reaction, was this the 0:05:02.640 --> 0:05:05.039 macro level event that we thought it was going to be. 0:05:07.200 --> 0:05:09.200 That's a great question. I would say that it's kind 0:05:09.200 --> 0:05:12.039 of close enough in line with expectations, even if it 0:05:12.120 --> 0:05:14.440 is a little bit shy the most optimistic ones that 0:05:14.480 --> 0:05:16.440 I don't think this is going to really cause people 0:05:16.440 --> 0:05:19.280 to really change how they're allocating, change their opinions on 0:05:19.680 --> 0:05:22.760 AI's as kind of fundamental secular driver. But maybe in 0:05:22.800 --> 0:05:24.920 the near term we do see a little bit of weakness. 0:05:24.960 --> 0:05:27.400 I mean, again, some of these stocks have been moving 0:05:27.480 --> 0:05:29.200 up so much there have been sort of kind of 0:05:29.240 --> 0:05:32.440 growing calls about their valuation. Some concerns about that. I 0:05:32.440 --> 0:05:35.080 don't know if this is the kind of absolute blowout 0:05:35.120 --> 0:05:37.400 that will just you know, cause people to continue piling 0:05:37.440 --> 0:05:40.200 into the the way that they were doing earlier this year. 0:05:41.120 --> 0:05:43.120 There is a lot in the news cycle inclusive of 0:05:43.200 --> 0:05:45.200 and outside of video. I think one of the names 0:05:45.200 --> 0:05:47.880 that you note in after ours is super Micro. It's 0:05:47.920 --> 0:05:52.240 down significantly. There's an video relationship to that, and then 0:05:52.279 --> 0:05:54.720 there's the news around that name in and of itself. 0:05:56.000 --> 0:05:58.960 Yeah. Absolutely so. We saw yesterday a short report came 0:05:59.000 --> 0:06:02.760 out Today's delaying the filing of the tin K. You know, 0:06:02.839 --> 0:06:04.800 both of those cars does a weakness in the stock. 0:06:04.839 --> 0:06:06.680 I think today it was down more than double digits, 0:06:06.760 --> 0:06:08.960 So you know, a lot of reasons to be concerned 0:06:08.960 --> 0:06:12.120 there in general. And it is one of Nvidia's biggest customers. 0:06:12.160 --> 0:06:14.320 I think it's the third largest, So you know, this 0:06:14.400 --> 0:06:17.160 is just another reason for people who might have been 0:06:17.240 --> 0:06:19.680 kind of souring on super Micro to you know, maybe 0:06:19.720 --> 0:06:20.799 be pulling the cell button. 0:06:20.640 --> 0:06:21.040 A little bit. 0:06:22.279 --> 0:06:25.159 Bloomdo's Ryan for Selica with the after hours action out 0:06:25.160 --> 0:06:27.719 of Chicago. Thank you. Let's get the reaction from the 0:06:27.760 --> 0:06:31.279 cell side with one of the stocks, relative bears d 0:06:31.360 --> 0:06:34.599 A Davidson's Gil Luria had a neutral rating and a 0:06:34.640 --> 0:06:38.000 street low ninety dollars price target on shares of Nvidia. 0:06:38.080 --> 0:06:41.920 He joins US now and Gil, I guess does that 0:06:42.080 --> 0:06:44.839 change your barished perspective on the stock and your main 0:06:44.920 --> 0:06:49.480 takeaway from the analyst school, the quarter was. 0:06:49.400 --> 0:06:53.400 A very good quarter there. Ability to continue to grow 0:06:53.920 --> 0:06:58.919 delivering chips at this rate at this scale is fantastic 0:06:58.960 --> 0:07:02.080 and unprecedented. I don't think there's much of a revenue 0:07:02.080 --> 0:07:04.960 issue here, of a growth issue here. I think a 0:07:05.000 --> 0:07:08.160 little bit of a pushback is probably more around the 0:07:08.200 --> 0:07:10.960 margin situation. You had a good conversation with Ian about 0:07:10.960 --> 0:07:13.920 Blackwell and some of the moving pieces part of what's 0:07:13.960 --> 0:07:17.800 happening with Blackwell. It's a new product, it's got a 0:07:17.840 --> 0:07:21.160 few hitches along the way that put pressure on gross margins, 0:07:21.440 --> 0:07:25.640 and then operating margins grew more than anticipated, so less 0:07:25.680 --> 0:07:28.800 of the top line upside flowed to the bottom line. 0:07:28.840 --> 0:07:31.840 And that's probably where investors are getting a little bit 0:07:31.880 --> 0:07:35.000 more cautious than they were before, because in the last 0:07:35.040 --> 0:07:39.240 several quarters, the huge upside to revenue flowed all the 0:07:39.240 --> 0:07:42.720 way to the bottom line, creating huge upside there to 0:07:43.160 --> 0:07:45.960 believe in and Nvidia right now is to believe twenty 0:07:46.040 --> 0:07:48.400 twenty six calendar is going to be about two hundred 0:07:48.440 --> 0:07:51.440 billion of revenue and about four and a half to 0:07:51.560 --> 0:07:55.320 five dollars of earnings. That's only possible if they continue 0:07:55.360 --> 0:07:58.160 to get this type of growth on the top line 0:07:58.280 --> 0:08:01.800 and more leverage. And the second part of that equation 0:08:01.880 --> 0:08:03.720 looks a little less certain right now. 0:08:04.280 --> 0:08:06.480 Gil. When you said that two hundred billion dollar figure, 0:08:06.520 --> 0:08:09.640 Bloombergsy and King, who's covered semi conductors at this company 0:08:09.920 --> 0:08:12.200 since nineteen ninety eight, sat next to me still, for 0:08:12.280 --> 0:08:15.480 what it's worth, gave a little smile. I mean, it's 0:08:15.480 --> 0:08:18.520 an astonishing figure. I think there's a lot of emphasis 0:08:18.560 --> 0:08:21.400 here on Blackwell. How did you react to the news 0:08:21.440 --> 0:08:24.960 of several billion dollars in fiscal fourth with Blackwell and 0:08:25.000 --> 0:08:29.360 the explanation that this was a production issue impacting the ramp, 0:08:29.440 --> 0:08:32.920 not a sort of core design issue around the product itself. 0:08:34.160 --> 0:08:35.960 So, first of all, our estimate is far less than 0:08:36.000 --> 0:08:38.520 two hundred billion for twenty twenty six. 0:08:38.440 --> 0:08:40.800 But that's you're a relative bear, Gill and I got 0:08:40.800 --> 0:08:41.360 that bit right. 0:08:41.400 --> 0:08:45.800 I think in terms of Blackwell, what happened to in 0:08:45.840 --> 0:08:47.319 n videos they used to be on a two year 0:08:47.360 --> 0:08:49.719 cycle for the data center product, and they decided to 0:08:49.760 --> 0:08:53.160 aggressively move to a one year cycle and then put 0:08:53.200 --> 0:08:56.120 a lot of new feature functionality and bundle into the 0:08:56.120 --> 0:09:00.200 Blackwell platform. So that was a very aggressive agenda, and 0:09:00.240 --> 0:09:03.520 it's not that surprising that there's some challenges that they 0:09:03.559 --> 0:09:07.480 have to encounter. If they can still ship in their 0:09:07.559 --> 0:09:11.439 fiscal fourth quarter, that's a very good sign. And by 0:09:11.480 --> 0:09:14.800 the way, their customers don't really care that much. Their 0:09:14.840 --> 0:09:17.200 customers are just going to continue to buy the latest 0:09:17.240 --> 0:09:19.640 and greatest, and if that's the Age two hundred, they'll 0:09:19.679 --> 0:09:22.120 just continue to buy h two hundred. They have made 0:09:22.120 --> 0:09:27.160 that abundantly clear on their conference calls, Microsoft, Meta, Amazon, Google, Tesla. 0:09:27.440 --> 0:09:30.400 Elon Musk made it abundantly clear they'll buy as much 0:09:30.440 --> 0:09:33.640 GPUs as and Video can produce, at least this year, 0:09:34.120 --> 0:09:36.679 because they're saying the demand signals are there. As long 0:09:36.679 --> 0:09:39.000 as the demand signals are there, they'll buy whatever and 0:09:39.120 --> 0:09:41.480 Video is selling, and towards the end of the year, 0:09:41.760 --> 0:09:43.280 a piece of that will be Blackwell. 0:09:44.520 --> 0:09:47.560 Gil Jensen Wong was it pains and gave a detailed 0:09:47.600 --> 0:09:51.440 answer that basically summarizing video as a systems vendor the 0:09:51.480 --> 0:09:55.920 discussion around Envy link, CPUs, everything beyond GPU. How much 0:09:56.000 --> 0:09:58.720 credit do you assign the company for that, the sort 0:09:58.760 --> 0:10:00.760 of complete systems that they sell. 0:10:01.880 --> 0:10:05.199 A tremendous amount of credit. Let's start with the beginning. 0:10:05.840 --> 0:10:10.040 When history is written, we'll talk about how Jensen Wangen 0:10:10.160 --> 0:10:14.199 Nvidia created the capabilities that make generative AI possible today 0:10:14.200 --> 0:10:16.640 in a similar way that Elon Musk is responsible for 0:10:17.000 --> 0:10:20.240 the electric vehicle. And part of the genius wasn't just 0:10:20.320 --> 0:10:23.600 seeing that this is coming, that accelerate computers coming, AI 0:10:23.720 --> 0:10:27.360 capabilities is coming well before anybody else did. It's also 0:10:27.480 --> 0:10:30.920 realizing that in order to create a moat around the chip, 0:10:30.960 --> 0:10:33.679 they need to do a couple of things. One is 0:10:34.080 --> 0:10:37.120 to own the proprietary software around that KUDA, but the 0:10:37.200 --> 0:10:40.920 other thing is to enhance the bundle as much as possible. 0:10:41.320 --> 0:10:45.280 These GPUs work in a raise when they work together, 0:10:45.520 --> 0:10:48.679 ten thousands or sometimes tens of thousands of GPUs together. 0:10:49.200 --> 0:10:55.439 When you box those up in a box with multiple GPUs, CPUs, wiring, 0:10:55.600 --> 0:10:59.600 cabling and firmware, you make it harder for your competitors 0:10:59.640 --> 0:11:02.040 to catch. And they've done a remarkable job of that. 0:11:03.120 --> 0:11:05.520 I also wonder about all the rest The story thus 0:11:05.559 --> 0:11:09.040 far has been about the hyperscale is a meta five names, 0:11:09.160 --> 0:11:12.720 essentially being the core customer of Nvidia. We got some 0:11:12.920 --> 0:11:16.120 detail on low double digit billions sales this year in 0:11:16.200 --> 0:11:19.560 sovereign AI. Do you see evidence that the market is 0:11:19.640 --> 0:11:22.119 broadening out beyond just the cloud providers? 0:11:23.320 --> 0:11:26.000 Not really. It seems like those five are still a 0:11:26.040 --> 0:11:30.640 substantial amount of the revenue. And those five have in 0:11:30.679 --> 0:11:33.680 their most recent calls said some interesting things. One of 0:11:33.720 --> 0:11:36.719 them is that they consistently said they are over investing. 0:11:37.120 --> 0:11:39.520 So they said they definitely want to invest more. We're 0:11:39.520 --> 0:11:41.520 seeing that in results today. But they also use the 0:11:41.600 --> 0:11:45.120 term over investing. Over investing is something you do before 0:11:45.520 --> 0:11:48.240 you do the other thing, and they made it clear 0:11:48.280 --> 0:11:50.400 that they're only going to do so when they have 0:11:50.480 --> 0:11:55.520 the demand signals. If for whatever reason, demand for AI 0:11:55.679 --> 0:12:01.160 capacity goes down, for instance, because compute doesn't continue to 0:12:01.280 --> 0:12:05.800 drive performance of foundational models, you're going to see those 0:12:05.840 --> 0:12:10.120 same five actors pull back and say, you know, maybe 0:12:10.120 --> 0:12:12.600 we have enough data center capacity, or maybe what we 0:12:12.679 --> 0:12:15.760 have in the pipeline is enough. And that's the concern 0:12:15.840 --> 0:12:19.079 for twenty twenty five and twenty six is that those 0:12:19.679 --> 0:12:23.480 very companies could do that. Beyond that, these are the 0:12:23.559 --> 0:12:27.160 same companies that are developing their own chips. They're at 0:12:27.240 --> 0:12:30.320 various stages of this, but Google's TPUs are probably as 0:12:30.360 --> 0:12:33.040 good as in videos, so in terms of their internal 0:12:33.120 --> 0:12:36.680 use and selling to customers such as Apple, Google's already 0:12:36.679 --> 0:12:41.560 caught up, Amazon's catching up, Meta and Microsoft are earlier stages. 0:12:42.120 --> 0:12:45.880 But these same companies that are buying every GPU from 0:12:46.040 --> 0:12:48.160 video they can gil will be the ones that will 0:12:48.200 --> 0:12:50.480 buy possibly less next year in the year after that. 0:12:51.240 --> 0:12:53.560 I just want to reflect for our TV and radio audience, 0:12:53.600 --> 0:12:55.960 there are people tuned in around the world. My understanding 0:12:56.040 --> 0:12:58.400 is there are watch parties in cities from New York 0:12:58.440 --> 0:13:01.920 to London. For this earnings. We've hyped it up for weeks. 0:13:02.360 --> 0:13:05.240 Probably this is the most unique earnings print that I've 0:13:05.280 --> 0:13:07.840 covered in my career. What about for you in the 0:13:07.880 --> 0:13:10.720 semiconductor space, Why is this so different? 0:13:11.720 --> 0:13:14.600 Well, so, first of all, I tried to participate by 0:13:14.640 --> 0:13:19.439 wearing a video green today. But importantly and video is 0:13:19.480 --> 0:13:22.719 important because of how much they've added to the cumulative 0:13:22.720 --> 0:13:25.880 earnings of this in P. Five hundred and the cumulative performance. 0:13:26.360 --> 0:13:31.559 But I wouldn't necessarily extrapolate their success or sometimes setbacks 0:13:31.800 --> 0:13:34.320 to the rest of the market because what they do 0:13:34.440 --> 0:13:38.480 is very specific. The growth and data center construction and 0:13:38.880 --> 0:13:45.200 data center a GPU fulfillment into those data centers is 0:13:45.240 --> 0:13:48.720 a very small set of companies. I'd argue, and Video 0:13:48.800 --> 0:13:51.280 is sucking the oxygen for the room from a lot 0:13:51.320 --> 0:13:54.880 of other technology companies. So we celebrate within Video because 0:13:54.880 --> 0:13:57.800 they've added so much value I wouldn't necessarily read through 0:13:57.880 --> 0:13:59.000 to the rest of technology. 0:14:00.080 --> 0:14:04.880 Guillerya of DA Davidson, the relatively most bearish or perhaps 0:14:04.960 --> 0:14:07.800 the least bullish name on the street. Thank you so 0:14:07.880 --> 0:14:11.600 welcome to this special edition of Bloomberg Technology for our 0:14:11.600 --> 0:14:14.360 TV and radio audience around the world. Joining me now 0:14:14.840 --> 0:14:18.920 is Jensen One, CEO of Nvidia, straight from the analyst's call, 0:14:18.960 --> 0:14:22.200 and Jensen, good evening, good afternoon to you. I think 0:14:22.280 --> 0:14:26.040 the market wanted more on Blackwell, they wanted more specifics, 0:14:26.080 --> 0:14:27.840 and I'm trying to go through all of the call 0:14:27.880 --> 0:14:31.560 and the transcript. It seems like a very clearly this 0:14:31.680 --> 0:14:34.480 was a production issue and not a fundamental design issue 0:14:34.480 --> 0:14:38.080 with Blackwell, but the deployment in the real world, what 0:14:38.160 --> 0:14:40.440 does that look like tangibly and is there a sort 0:14:40.440 --> 0:14:43.560 of delay in the timeline of that deployment and thus 0:14:43.640 --> 0:14:44.840 revenue from that product. 0:14:47.320 --> 0:14:53.320 I let's see, that's just the fact that I was 0:14:53.360 --> 0:14:56.840 so clear and it wasn't clear enough kind of tripped 0:14:56.880 --> 0:15:00.120 me up there right away. And so let's see. We 0:15:00.120 --> 0:15:03.320 we made a mass change to improve the yield. Functionality 0:15:03.320 --> 0:15:06.520 of Blackwell is wonderful. We're sampling Blackwell all over the 0:15:06.520 --> 0:15:12.800 world today. We show people giving tours to people of 0:15:12.840 --> 0:15:15.920 the Blackwall systems that we have up and running. You 0:15:15.960 --> 0:15:19.240 could find pictures of Blackwall systems all over the web. 0:15:20.240 --> 0:15:24.680 We have started volume production. Volume production will ship in 0:15:25.240 --> 0:15:29.240 Q four. Q four, we will have billions of dollars 0:15:29.280 --> 0:15:35.200 of Blackwell revenues and we will ramp from there. We 0:15:35.240 --> 0:15:40.120 will ramp from there. The demand for Blackwell far exceeds 0:15:40.280 --> 0:15:44.360 its supply, of course in the beginning, because the demand 0:15:44.400 --> 0:15:47.040 is so great, But we're going to have lots and 0:15:47.080 --> 0:15:50.480 lots of supply and we will be able to ramp. 0:15:51.040 --> 0:15:54.280 Starting in Q four, we have billions of dollars of 0:15:54.320 --> 0:15:57.040 revenues and we will ramp from there into Q one 0:15:57.120 --> 0:15:59.240 into Q two and two next year. We're going to 0:15:59.280 --> 0:16:00.000 have a great next year. 0:16:00.120 --> 0:16:04.760 Well, Jensen, what is the demand for accelerated computing beyond 0:16:04.760 --> 0:16:09.360 the hyperscalers and meta. 0:16:08.200 --> 0:16:13.000 Hyperscalers represent about forty five percent of our total data 0:16:13.000 --> 0:16:18.400 center business. We're relatively diversified today. We have hyperscalers, we 0:16:18.440 --> 0:16:24.720 have Internet service providers, we have sovereign AIS, we have 0:16:26.040 --> 0:16:32.680 industries enterprises, so it's fairly fairly diversified. A site outside 0:16:32.720 --> 0:16:38.760 of hyperscalers is the other fifty five percent. Now, the 0:16:38.840 --> 0:16:42.680 application use across all of that, all of that data 0:16:42.680 --> 0:16:48.560 center starts with accelerated computing. Accelerated computing does everything, of course, 0:16:49.040 --> 0:16:53.320 from well the models the things that we know about, 0:16:53.320 --> 0:16:57.400 which is generative AI, and that gets most of the attention. 0:16:58.120 --> 0:17:02.920 But at the core we also do database processing, pre 0:17:03.080 --> 0:17:08.119 and post processing of data before you use it for 0:17:08.280 --> 0:17:14.560 generative AI, trans coding, scientific simulations, computer graphics of course, 0:17:14.680 --> 0:17:18.760 image processing of course. And so there's tons of applications 0:17:18.800 --> 0:17:23.199 that people use are accelerated computing for, and one of 0:17:23.200 --> 0:17:26.880 them is generative AI. And so let's see what else 0:17:26.880 --> 0:17:29.480 can I say? I think that's the. 0:17:29.640 --> 0:17:32.639 Lever jump in Jensen please on Sovereign AI. You and 0:17:32.640 --> 0:17:35.400 I've talked about that before, and it was so interesting 0:17:35.440 --> 0:17:38.800 to hear something behind it that in this fiscal year 0:17:39.200 --> 0:17:41.000 there will be low double digit I think you said 0:17:41.040 --> 0:17:44.119 billions of dollars in sovereign AI sales. But to the 0:17:44.200 --> 0:17:46.960 lay person, what does that mean. It means deals with 0:17:47.040 --> 0:17:48.880 specific governments, if so, where. 0:17:50.480 --> 0:17:56.680 It's not necessarily Sometimes it deals with a particular regional 0:17:57.240 --> 0:18:01.119 service provider that's been funded by the government. And oftentimes 0:18:01.160 --> 0:18:03.520 that's the case in a case of in the case 0:18:03.520 --> 0:18:07.760 of Japan, for example, the the Japanese government came out 0:18:08.040 --> 0:18:14.320 and UH offered UH subsidies of a couple of billion dollars. 0:18:14.400 --> 0:18:19.480 I think for several different internet companies and telcos to 0:18:19.560 --> 0:18:24.320 be able to fund their AI infrastructure. UH India has 0:18:24.440 --> 0:18:30.880 a sovereign AI initiative going and they're building their AI infrastructure. Canada, 0:18:32.119 --> 0:18:40.920 the UK, France, Italy, missing somebody, Singapore, Malaysia, UH. You know, 0:18:41.040 --> 0:18:46.200 a large number of countries are subsidizing their regional data 0:18:46.200 --> 0:18:50.160 centers so that they could become able to build out 0:18:50.160 --> 0:18:56.640 their AI infrastructure. They recognize that their countries UH knowledge, 0:18:56.680 --> 0:19:00.760 their countries data digital data is also their natural resource, 0:19:01.800 --> 0:19:03.800 not just the land they're sitting on, not just the 0:19:03.840 --> 0:19:07.960 air above them. But they realize now that their digital 0:19:08.040 --> 0:19:12.760 knowledge is part of their natural and national resource, and 0:19:12.800 --> 0:19:16.720 they are to harvest that and process that and transform 0:19:16.760 --> 0:19:22.159 it into their national digital intelligence. And so this is 0:19:22.200 --> 0:19:24.480 what we call sovereign AI. You could imagine almost every 0:19:24.520 --> 0:19:29.320 single country in the world will eventually recognize this and 0:19:29.320 --> 0:19:30.840 build out their AI infrastructure. 0:19:31.640 --> 0:19:34.359 Jenson, you use the word resource, and that makes me 0:19:34.400 --> 0:19:37.320 think about the energy requirements here. I think on the 0:19:37.320 --> 0:19:40.200 cool you talk about how the next generation models will 0:19:40.240 --> 0:19:43.639 have many orders of magnitude greater compute needs. But how 0:19:43.640 --> 0:19:46.359 will the energy needs increase and what is the advantage 0:19:46.400 --> 0:19:51.159 you feel in Vidia has in that sense, Well. 0:19:51.040 --> 0:19:53.879 The most important thing that we do is increase the 0:19:54.040 --> 0:19:58.080 performance of and increase the performance and efficiency of our 0:19:58.119 --> 0:20:03.320 next generation. So Blackwell is many times more performance than 0:20:03.480 --> 0:20:07.120 Hopper at the same level of power used. And so 0:20:07.200 --> 0:20:11.280 that's energy efficiency, more performance with the same amount of 0:20:11.320 --> 0:20:14.640 power or same performance at a lower power And that's 0:20:14.720 --> 0:20:18.919 number one. And the second is using luca cooling. We 0:20:18.960 --> 0:20:21.880 support air cool we support air cooling, we support liqual cooling, 0:20:21.960 --> 0:20:24.240 but liqual cooling is a lot more energy efficient, and 0:20:24.320 --> 0:20:27.520 so the combination of all of that you're going to 0:20:27.520 --> 0:20:31.119 get a pretty large, pretty large step up. But the 0:20:31.160 --> 0:20:34.080 important thing to also realize is that AI doesn't really 0:20:34.119 --> 0:20:37.120 care where it goes to school, and so increasingly we're 0:20:37.160 --> 0:20:41.040 going to see AI be trained somewhere else, have that 0:20:41.080 --> 0:20:44.399 model come back and be used near the population, or 0:20:44.680 --> 0:20:47.720 even running on your PC or your phone. And so 0:20:47.720 --> 0:20:50.600 we're going to train large models. But the goal is 0:20:50.600 --> 0:20:53.040 not to run the large models necessarily all the time. 0:20:53.200 --> 0:20:55.000 You can, you can surely do that for some of 0:20:55.040 --> 0:21:00.280 the premium services and the very high value AIS, but 0:21:00.480 --> 0:21:02.960 it's very likely that these large models would then help 0:21:03.080 --> 0:21:06.320 to train and teach smaller models, and what we'll end 0:21:06.400 --> 0:21:10.879 up doing is have one large, few large models that 0:21:10.960 --> 0:21:13.200 are able to train a whole bunch of small models, 0:21:13.200 --> 0:21:14.200 and they run everywhere. 0:21:15.840 --> 0:21:20.800 Jensen, you explain clearly that demand to build generative AI 0:21:20.960 --> 0:21:24.000 product on models or even at the GPU level is 0:21:24.080 --> 0:21:28.080 greater than current supply. In black Wealth's case in particular, 0:21:28.320 --> 0:21:31.159 explain the supply dynamics to me for your products and 0:21:31.200 --> 0:21:34.440 whether you see an improvement sequentially quarter on quarter or 0:21:34.480 --> 0:21:36.640 at some point by the end of fiscal year into 0:21:36.680 --> 0:21:37.160 next year. 0:21:38.440 --> 0:21:41.720 Well, the fact that we're growing would suggest that our 0:21:41.840 --> 0:21:46.600 supply is improving and our supply chain is quite large. 0:21:46.960 --> 0:21:49.359 One of the largest supply chains in the world. We 0:21:49.440 --> 0:21:53.400 have incredible partners and they're doing a great job supporting 0:21:53.480 --> 0:21:56.320 us in our growth. As you know, we're one of 0:21:56.320 --> 0:22:00.199 the fastest growing technology companies in history. None of that 0:22:00.200 --> 0:22:03.600 would have been possible without very strong demand but also 0:22:03.720 --> 0:22:06.520 very strong supply. We're expecting Q three to have more 0:22:06.520 --> 0:22:08.840 supply than Q two, We're expecting Q four to have 0:22:08.880 --> 0:22:10.919 more supply than Q three, and we're expecting Q one 0:22:10.960 --> 0:22:12.919 to have more supply than Q four, And so I 0:22:12.920 --> 0:22:16.000 think our supply, our supply condition going into next year 0:22:16.080 --> 0:22:18.800 will be in will be a large improvement over this 0:22:19.000 --> 0:22:24.960 last year with respect to demand. Blackwell is just such 0:22:24.960 --> 0:22:28.639 a leap and there's several things that are happening, you know, 0:22:28.720 --> 0:22:31.919 just the Foundation model makers themselves. The size of the 0:22:31.920 --> 0:22:35.879 Foundation models are growing from hundreds of billions parameters to 0:22:36.080 --> 0:22:40.399 trillions of parameters. They're also learning more languages. Instead of 0:22:40.600 --> 0:22:43.560 just learning human language, they're learning the language of images 0:22:43.720 --> 0:22:49.080 and sounds and videos, and they're even learning the language 0:22:49.080 --> 0:22:52.119 of three D graphics. And whenever they are able to 0:22:52.160 --> 0:22:54.879 learn these languages, they can understand what they see, but 0:22:54.920 --> 0:22:58.119 they can also generate what they're asked to generate and 0:22:58.200 --> 0:23:02.400 so they're learning the language of routines and chemicals and physics. 0:23:02.960 --> 0:23:04.840